Cost Tracking

Cost Tracking

agent-os tracks token usage and costs for every agent run automatically. The SDK reports token counts and cost data with run_end events.

/costs — Cost dashboard

The cost dashboard provides a comprehensive overview:

30-day cost trend

A daily breakdown of your spending over the last 30 days.

Per-agent costs

See which agents are consuming the most tokens and budget.

Per-model costs

Compare costs across different LLM models (e.g., GPT-4 vs. Claude vs. Groq).

Token tracking

Each completed run records:

Field	Description
prompt_tokens	Tokens sent to the LLM
completion_tokens	Tokens generated by the LLM
total_tokens	Sum of prompt + completion tokens
cost_usd	Total cost in USD
model	Which LLM model was used

These values are extracted from the run_end event payload.

Budget warnings

When an agent has a budget set (see Agent Registry), the ingest API checks cost against budget after every run_end event. If the total cost exceeds the budget, the API response includes a budget warning:

{
  "status": "ok",
  "budget_warning": "Agent 'my-agent' has exceeded its budget of $10.00"
}

Your SDK or integration can check for this warning and take action (e.g., pause the agent).

Cost data API

The cost data endpoint (GET /costs/data) returns JSON suitable for charting:

• Daily cost totals (30-day window)
• Per-agent cost breakdown
• Per-model cost breakdown

This endpoint requires authentication.